THE CLIMATE AND ENERGY
ECONOMICS PROJECT
CLIMATE AND ENERGY ECONOMICS DISCUSSION PAPER | DECEMBER 4, 2014
DO GASOLINE PRICES AFFECT
RESIDENTIAL PROPERTY VALUES?
|
ADELE C. MORRIS
Brookings
HELEN R. NEILL
University of Nevada
Las Vegas
DO GASOLINE PRICES AFFECT
RESIDENTIAL PROPERTY VALUES? ∗
DECEMBER 4, 2014
ADELE C. MORRIS
The Brookings Institution
HELEN R. NEILL
University of Nevada, Las Vegas
∗
The authors thank Kristina Donathan, Danny Cohen, Nathan Joo, and Matthew Kinzer for able
research assistance. Edward Coulson, Ashley Langer, Joe Cortright, Justin Wolfers, Cliff Winston,
Gary Burtless, Ronald Cummings, and Jaewon Lim provided helpful suggestions, and the Clark
County Assessor’s Office provided the real estate data. Partial funding for this research was
provided by Brookings Mountain West and the Nevada NSF EPSCoR Infrastructure for Climate
Change Science, Education, and Outreach project (Cooperative Agreement EPS-0814372). The
paper reflects only the views of the authors and not their respective employers, the Brookings
Institution and the University of Nevada, Las Vegas.
EXECUTIVE SUMMARY
This paper estimates the effect of gasoline prices on home values and explores
the degree to which the relationship varies across a city. Using data from
930,702 home sales in Clark County, Nevada, from 1976 through 2010, we find
that gasoline prices have significantly different effects on the sales price of homes
in different neighborhoods. A ten percent increase in gasoline prices is
associated with changes in location-specific average home values that span a
range of over $13,000. This suggests that energy policies may affect household
housing wealth via gasoline prices, a heretofore unrecognized distributional
outcome.
1.
INTRODUCTION
A large literature explores the economic and distributional effects of potential
carbon pricing policies, including how taxing carbon could raise retail energy
prices and differentially affect households across regions and income classes. 1 A
growing literature connects the effects of energy prices to the value of durable
household goods. 2 For example, Busse et al (2013) estimate consumers’
willingness to pay for vehicle fuel economy and find that a $1 increase in the
price of gasoline is associated with an increase of $354 in the average price of
the highest fuel economy quartile of cars relative to that of the lowest fuel
economy quartile. They found an estimated relative price difference of $1,945
for used cars.
Hedonic studies of home values have analyzed how home heating oil prices and
energy efficiency investments could be capitalized into home values, 3 but so far
1
EPA (2010), Hassett et al (2009), Rausch et al (2010), Mathur and Morris (2014).
Langer and Miller (2013)
3
Walls et al (2013) review this literature and find that households do value the energy-related
costs of home ownership as reflected in the price premium associated with energy efficiency
2
1
no research has examined the potential of fuel taxes to affect home prices, and
little research has related gasoline prices to households’ willingness to pay for
homes. 4 Households adjust to changes in gasoline prices via their choices of
vehicle, where to live and work, and mode of transport. Communities adjust
through investments in infrastructure and housing supply. Inducing these shifts is
key to reducing greenhouse gas emissions from transport fuels. Thus,
understanding how housing markets respond to gasoline prices is important for
anticipating the environmental and distributional outcomes of energy policies.
The relationship between gasoline and housing markets could also be important
to understanding broader housing market dynamics, including the relative prices
of close-in and far flung homes, mortgage performance, the 2008 housing bust,
and the location of new housing supply. For example, Glaeser et al (2012) find
that during the last two housing booms, home prices rose significantly less in
areas farther from the central business district. Gasoline prices could help
explain why their result was far more pronounced during the 1996-2006 housing
boom (a period in which real gasoline prices tripled) than during 1982-89
housing boom (a period in which real gasoline prices fell) and why, during the
second boom, the price growth gradient was flatter in areas in which more
adults take public transit.
Some evidence links gasoline prices to lending risks and the 2008 housing bust.
For example, Kaufmann et al (2011) show that spikes in gasoline prices can
stress household budgets and contribute to mortgage non-performance.
Cortright (2008) relates the initial signs of the collapse of the housing bubble
with the run-up in oil prices in 2008. And Sexton et al (2012) connect the rapid
run-up of gasoline prices in 2007 to 2008 to the bursting of the U.S. housing
market bubble. Their simulations support their theory that the greater the
commuting distance from the central business district, the more a gasoline price
shock causes home values to decline.
certification. Eicholtz et al (2013) find that certified green buildings have rents and asset prices
that are significantly higher than those documented for conventional office space.
4
Coulson and Engle (1987) find that increases in gas prices between 1974 and 1979 boosted the
differential between central city and suburban house prices in a sample of six cities.
2
Molloy and Shan (2013) investigate the effect of gasoline prices on the location of
new home construction and home prices across the United States. They find
that a 10 percent increase in gasoline prices leads to a 10 percent decrease in
construction in locations with a long average commute (greater than 24 minutes)
relative other locations, but they find no significant change on house prices.
However, they find that in areas where the housing supply is constrained by
regulatory or geographic factors and demand for housing is growing, gasoline
prices have a significant negative effect on home prices in areas with long
commutes. Their results suggest that if gasoline prices fall and demand for homes
in the suburbs grows, developers usually just build more suburban homes. If they
can’t for some reason, then the value of those suburban homes goes up, and the
lower gasoline prices can boost suburban home values.
The paper extends the literature in several ways. It is the first to look for direct
evidence of gasoline prices on property values using property-level data and the
first to directly estimate price-price elasticities of the two goods or to
characterize the question in that way. This paper is also the first to produce
graduated color maps that show spatial variations in the gasoline price effects,
their statistical significance, and the estimated mean effects on home values. We
find that the elasticities are statistically significantly different from each other in
most of the more than 50,000 location-location pairwise F-tests, and we graph
the results in a way that is common in science but new to the economics
literature. 5 Finally, we directly apply our results to the potential outcomes of
climate and energy policies.
Our approach is different than that of Molloy and Shan (2013) in that our data is
more spatially detailed and allows estimation of within-city differences in gasoline
price effects with no a priori assumptions about which neighborhoods are likely
to be most sensitive to gasoline prices (such as those with long commutes). This
allows us to pick up relationships that could depend on factors other than workrelated travel.
As noted by Malloy and Shan (2013), significant capitalization of gasoline prices
may be more likely in areas where short run housing supply is inelastic, such as
5
Padmos et al (2008), Table 1; Araya et al (2014), Extended Data Figure 10: Full-resolution view
of global pairwise transcription factor co-association matrix.
3
Las Vegas. According to Saiz (2010), about 32 percent of the Las Vegas
metropolitan area is undevelopable owing to physical or regulatory constraints,
making it the 32nd most-constrained housing supply in the United States of the 95
metropolitan statistical areas that have a population over 500,000. In addition,
the U.S. federal government controls about 90 percent (about 4.66 million acres)
of the land in Clark County. 6 Several land sales by the U.S. Bureau of Land
Management expanded the land available for housing development during the
period of our data, but the large share of government ownership and the
mountainous terrain surrounding the Las Vegas valley may, at least at times and
in some places, impede the response of private markets to increased housing
demand. 7 According to Saiz (2010), Las Vegas is less constrained than a number
of other large metropolitan areas such as New York City, Chicago, San Francisco,
Los Angeles, San Diego, Miami, and Boston. Thus our results could suggest a role
for gasoline prices in home values in other major property markets.
This paper proceeds as follows. Section 2 describes the hedonic model. Section
3 describes the data for property sales and gasoline prices. It also addresses
potential problems of misspecification. Section 4 reviews the results, and Section
5 concludes. The appendix describes the treatment of the data in more detail
and presents the results of an alternative (repeat sales) model.
2.
METHODOLOGY
Hedonic regression is a standard approach for estimating the relationship
between the prices of houses and their characteristics. 8 Owing to its intrinsically
reduced form nature, no particular theory governs the structure or functional
form of hedonic price models. A hedonic price schedule is simply the locus of
tangencies between consumers’ bid functions and suppliers’ offer functions. 9
Our standard hedonic property price model postulates that the inflation-adjusted
selling price of a house is a function of its physical characteristics and other
6
Clark County Department of Comprehensive Planning (2013).
The Southern Nevada Public Land Management Act of 1998 allows the Bureau of Land
Management to sell public land within a specific boundary around Las Vegas.
8
Rosen (1974), Campbell et al (2011)
9
Chay and Greenstone (2005)
7
4
factors. The data include 930,702 home sales in Clark County, Nevada, from
1976 through 2010. Clark County has a population of over two million residents
and includes the two largest cities in Nevada, Las Vegas and Henderson. Its
sprawling metropolitan area epitomizes large automobile-dependent western-U.S.
cities.
The typical log-log hedonic estimation equation for yist, the log price of house
i in census tract s in the period of sale t, is:
1)
yist = c + αs + γt + β′ Xi + λsgmαs + μsumαs + ϵist.
The variable c is a constant. The terms αs are indicators for the census tracts in
which the properties lie. Alone, they control for time-invariant characteristics of
neighborhoods. Most census tracts are small, so in general the tract indicators
do a good job controlling for location; the median census tract is only .71 square
miles in area. Tract areas range from 0.16 to 2,181 square miles, and ten of our
224 census tracts have land areas over 100 square miles. Despite the large size
of some of the tracts, the observations are clustered close to the Las Vegas
metropolitan area, and the terms αs are good indicators for location. The
appendix maps the observations and reports the results of a repeat sales
approach that controls even more carefully for location.
The terms γt are year indicators, and they control for broad trends in the Clark
County housing market. 10 Period m (the month of sale) falls in year t. The vector
Xi includes property characteristics that are standard in hedonic models, with
coefficients β.
The variable gm is the log price of gasoline in the month in which the home is
sold, and the gmαs terms are its interactions with the census tract indicators.
Thus the coefficient λs is the elasticity of the price of homes in census tract s with
respect to the price of gasoline. Our focus is on the variation in and spatial
patterns of these estimated elasticities. We expect λs will be relatively higher in
locations that become more attractive when gasoline prices rise (all else equal)
and λs will be relatively lower in locations that are less attractive when gasoline
10
Including the interaction of census tract and year indicators is infeasible given the large number
of years (35) and tracts (334) in our data.
5
prices rise. The absolute value of λs will be larger in areas most sensitive to
gasoline prices, and it will be closer to zero in areas that are insensitive to
gasoline prices. Equation 1 imposes no a priori restrictions the gasolinedependent location premia (or discounts) in each tract, such as a particular
functional relationship with the distance from central business districts. This
function-free approach is appropriate to Clark County, which has multiple major
employment centers.
The terms μsumαs help control for macroeconomic fluctuations that may affect
home values differently in different neighborhoods. Broad macroeconomic
fluctuations could drive both housing demand and oil demand, and that will shift
the distribution of our estimated elasticities. By itself, that is not a problem for
our study because we are focused on the relative value of homes in different
neighborhoods, not area-wide average property values. However, in principle a
strong economy could result in higher demand for housing in some
neighborhoods than others. Likewise, a weak economy could depress the value
of homes in some neighborhoods more than others, while also being correlated
with lower gasoline prices. If this occurs, then the estimated coefficients on the
location-gasoline price interaction terms could be biased. Thus, we include the
terms um, measures of the unemployment rate in Nevada in the month of the
property sale, interacted with the location indicators αs. The variable ϵist is an
error term that reflects random variation in house prices.
Some hedonic studies, particularly those that construct housing price indices,
must carefully control for the physical characteristics of homes that are sold to
reveal the underlying city-wide trends in home values as an asset class. 11 In our
application, as long as we appropriately control for the location of the home, the
results of interest are not sensitive to whether the quality of homes or other
unobserved characteristics of homes sold vary over time.
It may be the case that when gasoline prices rise, more homes further from the
city center are offered for sale while buyer interest in those distant homes
wanes. Thus the volume of sales in those areas could on net go up or down, but
11
Bajari et al (2012) review the issues that arise when omitted attributes are correlated with the
observed variables. See Dorsey et al (2010) for more on estimating housing price indices.
6
either way, prices should unambiguously fall. The hedonic model could fit
transactions in one neighborhood in a year better than another neighborhood in
that same year. If so, errors would not be independent and identically distributed;
the OLS estimates would be unbiased but the standard errors would be wrong.
To address this, we estimate robust standard errors, and we cluster the errors
by tract-year. 12
3.
DATA
This study combines data on property transactions with data on gasoline prices.
In addition, we use data on the unemployment rates in Nevada from the U.S.
Bureau of Labor Statistics.
A.
Property data
We use data from about 931,000 arms’ length home sales in Clark County,
Nevada, from January 1976 to December 2010, obtained from the Clark County
Assessor’s Office. Clark County is an extreme example of the boom and bust
U.S. housing market experience, and real home prices exhibit strong variation.
The population of Clark County grew dramatically over the duration of the data,
with an average annual growth rate of 5 percent from 1990 to 2009. The county
population more than doubled from about 780,000 in 1990 to more than two
million in 2009. 13 To illustrate the broad market context of our study, Figure 1
shows the S&P/Case-Shiller Home Sales Price Indices for Las Vegas and a
broader Case-Shiller composite index for ten large U.S. cities. The boom was
steeper and the freefall in 2008 was more dramatic in Las Vegas than in most
other cities.
12
Nichols and Schaffer (2007)
Population data downloaded from Clark County Assessor on September 22, 2010, from
http://www.accessclarkcounty.com/depts/comprehensive_planning/demographics/Pages/Demogra
phics.aspx.
13
7
FIGURE 1. S&P/CASE-SHILLER HOME PRICE INDICES, JANUARY 1990 TO DECEMBER
2010
250
200
150
100
50
0
Las Vegas
Composite of 10 Cities
Notes: Data for this figure are the S&P/Case-Shiller Seasonally Adjusted Home Price
Index Levels, downloaded August 2, 2011, from
http://www.standardandpoors.com/indices/sp-case-shiller-home-priceindices/en/us/?indexId=spusa-cashpidff--p-us----.
Each unit of observation in the data is the sale of a residential property. The data
include the actual selling price of the property, the sale date, and detailed
characteristics of the home. As shown in TABLE 1, the property characteristics
include lot size, square footage of living area, number of full baths, the age of the
home at the time of sale, and indicators for amenities such as a pool.
8
TABLE 1.−SUMMARY STATISTICS
Number of Observations = 930,702
Mean
Property price ($)
Living space (ft2)
Lot size (acres)
Home age (years)
Pool indicator
Full bathrooms
Foreclosure indicator
Townhouse indicator
Multiplex indicator
Price per gallon of gasoline
($)
Nevada state-level
unemployment rate (%)
245,019
1,904
.16
9.32
.21
2.19
.063
0.07
0.01
Standard
Deviation
146,315
759
.15
12.48
.41
.58
.244
0.26
0.09
Min
Max
40,005
280
.01
0
0
1
0
0
0
2,309,211
7,988
5
109
1
6
1
1
1
2.18
0.62
1.22
4.08
6.44
3.16
3.8
14.9
Notes: All dollar values appear in 2010 dollars, deflated using the Consumer Price
Index for All Urban Consumers. The data set combines information from the Clark
County Assessor’s Office, the U.S. Census, and the U.S. Energy Information
Administration.
We also include an indicator for transactions the Clark County Assessor
designates as linked to a foreclosure. Campbell et al (2011) note that illiquidity in
the housing market can depress prices of forced sales. Using data from over
1,800,000 home sales in Massachusetts from 1987 to March 2009, they find
foreclosed homes sold at substantial discounts relative to other sales, about 27
percent less on average. We therefore include an indicator in this study to
estimate a similar average discount for the Clark County foreclosure
transactions from 1976 through 2010.
We removed outliers from the data as detailed in the Appendix. For example,
given the specialized market for such properties, we exclude very large homes
(greater than 8,000 sq. ft. of living area) and homes on very large lots (greater
than five acres). We also drop sales of properties with prices under $40,000 or
9
over $5 million ($2010). These dropped observations collectively represent 4.6
percent of the raw dataset.
FIGURE 2 shows the distribution of the remaining 930,702 observations by year of
sale from 1976 to 2010, with sales of new homes in blue and existing homes in
red. The observations are heavily weighted towards the past two decades for
two reasons. First, new home sales grew dramatically from the mid-1990s
through 2006, as evidenced by the heights of the blue bars. Second, the Clark
County database includes only the three most recent transactions for each
property. Thus, although we observe some sales back to 1976, those
observations are only of homes that were not subsequently sold more than two
additional times by the end of 2010.
FIGURE 2. NUMBER OF OBSERVATIONS BY YEAR OF SALE, NEW AND EXISTING
HOMES
80,000
60,000
40,000
0
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
2008
2010
20,000
Existing
B.
New
Gasoline prices
Many hedonic studies infer the implicit price function for non-market factors
such as air quality and natural open space. In that context, consistent estimation
is difficult because unobserved factors (such as crime rates) could covary with
10
both the non-market factors and housing prices. 14 Here, one might be concerned
that, say, seasonality in gasoline prices and seasonality in housing markets could
produce spurious correlation and compromise the interpretation of our
estimated coefficients. Gasoline prices show the expected seasonal pattern of
systematically higher levels in summer months, but property sales prices in Clark
County, Nevada, do not show a strong seasonality -- less than one percent
variation in sales price by calendar month of sale. The climate, the large share of
new home sales in the data, and the attractiveness of Las Vegas for retirement
and second homes could all contribute to this. 15
Potential misspecifications could also arise with local gasoline prices, which are a
function of specific neighborhood characteristics, such as the scale and
competitiveness of the local retail economy. We avoid this problem by using
national average gasoline prices, which are unaffected by neighborhood
characteristics in Clark County, Nevada. Another reason to use national gasoline
prices in this study is that a key goal of this paper is to understand how national
climate and energy policies can affect local housing markets. In that context it
makes sense to exclude within-city gasoline price deviations, which are unlikely
to be affected by national or state-level energy policy. FIGURE 3 shows the U.S.
national average of gasoline prices from the U.S. Energy Information
Administration (EIA) in nominal and 2010 dollars. The figure shows significant
variation in real gasoline prices over the period of this study, from a low of $1.22
in February 1999 to a high of $4.08 per gallon in June 2008.
14
For example, see Gayer et al (2002). Chay and Greenstone (2005), Halvorsen and Pollakowski
(1981), and Cropper et al (1988) discuss the challenges of misspecification of hedonic price
models.
15
We estimated Equation (1) both with and without calendar month indicators and find that they
add no appreciable explanatory value.
11
FIGURE 3. MONTHLY U.S. AVERAGE MOTOR GASOLINE RETAIL PRICE, JANUARY 1976
TO DECEMBER 2010
Notes: Data for gasoline prices are from EIA’s Short-Term Energy Outlook - Real
Energy Prices data series, February 2011 update. See
http://www.eia.gov/EMEU/steo/realprices/index.cfm. Prices are for regular grade gasoline.
Households should consider the expected price of gas when deciding how far to
live from work. Empirically, however, the current level of gas prices appears to
be a good proxy for the expected future price. 16 Anderson et al (2011) find that
the average consumer expects the future real price of gasoline to equal the
current price, and consumers exhibit a reasonable forecast in most instances.
4.
RESULTS
A. Hedonic Model
Table 2 reports the estimation results of a model that includes log gas prices and
Nevada unemployment rates, but not their interactions with location indicators.
16
Molloy and Shan (2010), Alquist and Kilian (2010), Bopp and Lady (1991), and Chinn and
Coibion (2010)
12
TABLE 2.− MODEL WITHOUT INTERACTIONS
DEPENDENT VARIABLE: LN(PROPERTY SALE PRICE)
Explanatory
Variable
Estimate
Robust
Std.
Err.
t
P>|t|
Constant
8.288
*** 0.0502
165.03
0.000
ln(gasoline price)
0.051
*** 0.0075
6.71
0.000
Nevada
unemployment
-0.049
*** 0.0014
-33.68
0.000
rate (in percent)
ln (Square
0.579
*** 0.0051
112.65
0.000
footage of home)
ln(Lot size in
0.141
*** 0.0040
35.50
0.000
acres)
Age
-0.007
*** 0.0003
-23.09
0.000
Age*Age
0.000
0.0000
-0.39
0.697
Pool indicator
0.066
*** 0.0012
54.02
0.000
Number of full
0.027
*** 0.0018
15.31
0.000
baths
Multiplex
-0.032
*** 0.0088
-3.60
0.000
indicator
Townhouse
0.000
0.0049
0.05
0.957
indicator
Foreclosure
-0.147
*** 0.0047
-31.51
0.000
indicator
Year indicators (shown in Figure 4)
Census tract indicators (count = 335)
Census tract indicators interacted with ln(gasoline price)
Census tract indicators interacted with Nevada unemployment
rate in sale month
Number of observations =
R-squared =
Root MSE =
95% Conf. Interval
8.1893
0.0358
8.3862
0.0653
-0.0515
-0.0458
0.5693
0.5894
0.1327
0.1483
-0.0081
0.0000
0.0639
-0.0068
0.0000
0.0687
0.0238
0.0308
-0.0488
-0.0144
-0.0094
0.0100
-0.1560
-0.1377
Yes
Yes
No
No
930702
0.810
.227
Asterisks ***, **, and * indicate statistical significance at the 1%, 5%, and 10% levels,
respectively. Dollar values are in real $2010. Standard errors are clustered by tract-year,
with a total of 9759 distinct groups.
13
The coefficient in Table 2 on the log of gasoline prices suggests that a one
percent increase in the real gasoline price is associated with a .05 percent
increase in home values. This likely reflects a small broad positive relationship
between gasoline prices and home prices via macroeconomic conditions, even
when controlling for state unemployment rates. In other words, when economic
growth is strong, so are both home values and demand for transportation fuel. A
spatially systematic variation in this correlation could lead to spurious results
when we interact gasoline prices with location indicators. For example, home
values in the urban core could be relatively more closely tied to a cyclical tourist
economy than homes on the outskirts of town, and thus be more positively
related to gasoline prices. Ideally, we would employ an instrument for gasoline
prices that is uncorrelated with macroeconomic conditions, but such an
instrument is elusive. Rather we simply note that the magnitude and distribution
of the estimated elasticities may be skewed if home values in some
neighborhoods are more procyclical than in others, even when controlling for
state-wide macroeconomic conditions.
The other estimated coefficients on the housing characteristics shown in Table 2
are broadly consistent with other hedonic studies such as Walls et al (2013). For
example, the results suggest that a one percent increase in the square footage of
the home, all other factors equal, translates into about a 0.6 percent increase in
the sales price of the property. The estimated coefficient on the foreclosure
indicator suggests that distressed property sales produce a 13.5 percent lower
selling price than other sales of comparable properties. Our result is smaller but
of the same order of magnitude as the results of Campbell et al (2011), who
estimated foreclosure discounts on average of 27 percent on 1.8 million house
transactions in Massachusetts from 1987 through March 2009. Figure 4 shows
the year indicators in the model in Table 2. From 1990 on, they track the CaseShiller indices in Figure 1, with the exception that the year indicators in our
model show a dip in the mid-1990s.
14
FIGURE 4. YEAR-SPECIFIC EFFECTS IN MODEL IN TABLE 2, 1976 TO 2010
0.6
0.5
0.4
0.3
0.2
0.1
0
-0.11976
1980
1984
1988
1992
1996
2000
2004
2008
-0.2
-0.3
Notes: This graph reports the year specific effects in the model reported in Table 2.
The values on the vertical axis indicate the overall change in real prices of properties
sold relative to 1976.
As discussed above, the variance of the error term may not be constant for a
number of reasons, including because the fit of the model could be systematically
worse for certain property sub-markets than others. For example, Goodman and
Thibodeau (1997) demonstrate strong heteroscedasticity in hedonic housing
price models that is related the age of the dwelling. We conducted a Breusch–
Pagan test, which confirmed heteroscedasticity. We estimate robust standard
errors and cluster the errors by tract-year.
Table 3 reports the estimation results of our core model, which includes
interactions between the price of gasoline and state unemployment rates with
location indicators. The coefficients on the hedonic variables show little
difference from the model in Table 2.
15
TABLE 3.−MAIN RESULTS: MODEL WITH LOCATION INTERACTIONS
DEPENDENT VARIABLE: LN(PROPERTY SALE PRICE)
Explanatory
Variable
Robust
Std. Err.
Estimate
t
P>|t|
95% Conf. Interval
Constant
8.512
***
0.064
131.9
0.000
8.3856
8.6386
ln (Square footage
of home)
ln(Lot size in
acres)
0.577
***
0.005
113.86
0.000
0.5670
0.5869
0.144
***
0.004
36.62
0.000
0.1366
0.1520
-0.009
***
0.000
-30.48
0.000
-0.0094
-0.0083
Age*Age
0.000
***
0.000
6.18
0.000
0.0000
0.0000
Pool indicator
0.068
***
0.001
55.09
0.000
0.0652
0.0700
Number of full
baths
0.027
***
0.002
15.46
0.000
0.0235
0.0303
-0.021
***
0.008
-2.58
0.0100
-0.0375
-0.0051
0.659
-0.0074
0.0117
0.000
-0.1420
-0.1262
Age
Multiplex indicator
Townhouse
0.002
0.005
0.44
*
indicator
Foreclosure
-0.134
0.004
-33.23
***
indicator
Year indicators
Census tract indicators (count = 335)
Census tract indicators interacted with ln(gasoline price)
Census tract indicators interacted with Nevada unemployment
rate in sale month
Number of observations =
R-squared =
Root MSE =
Yes
Yes
Yes
Yes
930,702
0.819
0.222
Notes: Asterisks ***, **, and * indicate statistical significance at the 1%, 5%, and 10%
levels, respectively. Dollar values are in real $2010. Standard errors are clustered by
tract-year, with 9758 distinct groups.
16
Tables 4 and 5 report summary statistics on the estimated elasticities, and Figure
5 is a histogram of the values. The elasticities of home prices with respect to
gasoline prices range across the set of 335 census tracts from -.77 to 1.09, with
an (un-weighted) mean value of .06.
TABLE 4.−DESCRIPTIVE STATISTICS OF ESTIMATED ELASTICITIES
Number of Observations = 335
Variable
Interaction of Census
Tract indictors and
Property price
Interaction of Census
Tract indictors and
Nevada state-level
unemployment rate (%)
Mean
Standard
Deviation
Min
Max
0.0607
0.1388
-0.7731
1.0854
-0.0511
0.0193
-0.0905
0.0246
TABLE 5. − CENTILES OF 335 ESTIMATED GASOLINE PRICE ELASTICITIES
Percentile
Centile
Value for
Elasticity
10
20
30
40
50
60
70
80
90
-0.078
-0.014
0.023
0.051
0.066
0.086
0.107
0.134
0.183
17
FIGURE 5. HISTOGRAM OF ESTIMATED GAS PRICE ELASTICITIES IN 335 CENSUS TRACTS:
HEDONIC MODEL
A few of the estimated coefficients in Figure 5 have implausibly large absolute
values. For example, three tracts have estimated elasticities in excess of .5,
meaning that a 10 percent increase in the price of gasoline is associated with an
increase the home value of 5 percent. 17 Another tract recorded an implausibly
low elasticity of -0.773. 18 In most of these cases, a large share of a relatively low
number of tract-level observations occurred in a one or two-year timeframe. 19
Such temporally concentrated observations can produce anomalous coefficients
as a result of coincidental moves in gas prices and home values.
Abstracting from implausible tail estimates, we find that the 10th and 90th
percentiles of the estimated elasticities are -0.08 and 0.18, respectively. This
means that a ten percent increase in the price of gasoline is associated with a
17
These elasticities are 0.506 for tract 2962, 0.625 for tract 5609, and 1.085 for tract 2602.
Tract 2506
19
To illustrate, 27.55 percent of tract 2602’s 98 home sales occurred in 2007. 25.8 percent of
the population of only 310 home sales that took place in Tract 5609 occurred in 2003-2004.
18
18
range of home price changes between about negative one percent to positive
two percent.
The estimated elasticities for each of the tracts are illustrated in the graduated
color maps in Figure 6 and Figure 7.
FIGURE 6. ESTIMATED ELASTICITIES OF PROPERTY PRICES WITH RESPECT TO
GASOLINE PRICES: CLARK COUNTY, NEVADA
19
FIGURE 7. HEDONIC MODEL’S ESTIMATED ELASTICITIES OF PROPERTY PRICES WITH
RESPECT TO GASOLINE PRICES: CENTRAL CLARK COUNTY (LAS VEGAS, NORTH LAS
VEGAS, AND HENDERSON)
Central
Las Vegas
The color of each tract is coded to its estimated elasticity of property prices
with respect to gasoline prices, with each color comprising ten percent of the
tracts. Major highways appear in red. Figure 7 reports the same information as
Figure 6, zooming in on central Clark County. The red star in Figure 7 is an
intersection of major highways labeled “central Las Vegas.” The Las Vegas strip
lies on a diagonal directly below that point towards the southwest. White areas
are tracts in which there are no property transactions in our data, and they
include McCarran International Airport, the University of Nevada Las Vegas, and
areas dense in hotels and casinos.
The figures and maps show a wide range of estimated elasticities. Some areas of
higher elasticities (darker areas) appear in the urban core of Las Vegas and some
20
higher elasticities ring the city along major highways. There are also darker
patches near other cities within Clark County, such as Henderson and North Las
Vegas. Figure 8 maps the location of these cities.
FIGURE 8. CENTRAL CLARK COUNTY: LAS VEGAS, NORTH LAS VEGAS, AND
HENDERSON 20
Figure 7 shows a notably dark strip of relatively high and positive elasticities
running from the southwest to the northeast diagonally through the Las Vegas
strip area. Mean elasticities appear to fall in distance from the strip, and
elasticities in rural areas are generally small or negative, as shown by the lightest
20
Las Vegas, Google Maps, Downloaded February 7, 2014.
21
blue areas. The color pattern is broadly consistent with the hypothesis that the
relative values of properties in areas closer to the city center are more positively
correlated to gasoline prices, but the map also indicates some differences
between estimated elasticities even across adjacent tracts.
A few of the very rural tracts report relatively high elasticities, as shown by their
dark blue color. Some of these tracts have very few observations, and a number
of these elasticities, although relatively large, are not statistically different from
zero at a 90 percent confidence interval. The pattern of declining elasticity from
the urban core is even more pronounced in the repeat sales results, discussed in
the Appendix, than it is in Figure 7. This is consistent with the more recent
profile of sales in the repeat sales dataset.
FIGURE 9 reports the P-value of the estimated elasticities. The white tracts are
those with an estimated elasticity that is not statistically significantly different
from zero at the 90 percent confidence level or lower, meaning that one cannot
reject the hypothesis that gasoline has no effect on home values. The darkest
blue and red tracts are those in which we reject the null that the elasticity is
zero with a 99 percent or higher degree of confidence with positive and negative
elasticities, respectively. The two intermediate colors (orange and yellow for
negative elasticities; light blue and green for positive elasticities) represent
confidence levels of 95 to 99 percent and 90 to 95 percent, respectively.
22
FIGURE 9. P-VALUES OF ESTIMATED GASOLINE PRICE ELASTICITIES
PANEL A
PANEL B
23
The maps suggest that the results with the greatest (positive) statistical
significance, the dark blue areas, are those closest to the city center and to the
west and south of Las Vegas around the Highway 215 corridor. The most
statistically significant and negative elasticities are one large tract to the south of
the metropolitan area and a cluster to the west, which is surrounded by tracts
with negative and highly statistically significant elasticities.
Owing to small number of transactions, some of the very rural and mid-city
tracts do not show a statistically significant relationship between gasoline prices
and home values despite having relatively large estimated elasticities in absolute
value.
The spatial pattern of the elasticities is partly consistent with the theory that the
values of closer-in homes tend to be more positively related to gasoline prices
than further-out homes within the same housing market, but the effect is far
from a simple monotonic relationship with distance from the urban core. The
confidence around the results is particularly strong for central and suburban
areas, but elasticities in adjacent areas do not necessarily have the same sign or
significance.
The results described so far indicate that gasoline prices do indeed affect home
values in a significant share of neighborhoods across the county. We now turn to
question of the extent to which the estimated elasticities in different tracts are
significantly different from each other. To get at that, we conducted a full set of
55,945 unique pairwise F tests, each one testing a hypothesis that a particular
estimated elasticity is equal to another. 21 The color-coded results appear in a
(necessarily) symmetric 335 by 335 matrix in Figure 10. As a simple organizing
format for the matrix (but not pertinent to the econometric approach), we
calculated the mean longitude and latitude coordinates for the observations
within each tract (i.e., the “center” of the tract). Then we sorted the tracts by
the distance of their center to central Las Vegas. Thus, the upper left hand
corner of the matrix reports the P-values of the center-city tracts compared to
each other and the lower right shows the results of the most rural tracts
21
The number of unique pairs of census tracts is given by (335 x 334)/2 = 55,945.
24
compared to each other. The P-values of the pairwise F-tests fall into blue
confidence/color categories similar to the map in Figure 9, with the darkest
elements representing a 99 percent confidence level that the estimated
elasticities are statistically significantly different from each other and the white
elements representing a confidence level below 90 percent.
FIGURE 10. PAIRWISE F-TEST P-VALUES OF ESTIMATED GASOLINE PRICE ELASTICITIES
Legend
Range of P Value of F test
P > .1
.1 ≤ P < .05
.05 ≤ P < .01
P < .01
Fill Color
Notes: Rows and columns are 335 census tracts sorted by distance to central Las
Vegas. Thus the upper left tests tracts close to central Las Vegas against each other, and
the lower right tests exurban tracts against each other.
25
The upper right and lower left areas of the Figure 10 matrix show the urban
tracts compared to the rural tracts. Those areas appear generally darker,
informally supporting the intuition that property values in the urban core and
rural/exurban areas tend to have significantly different relationships with gasoline
prices. But the figure also shows that significant differences arise across
neighborhoods equidistant from the urban core.
B. Elasticities in Sub-Periods
The relationship between home prices and gasoline prices may have changed
through the 35 year period of our data. Certainly, the spatial pattern of
employment centers and housing developments has shifted significantly in Las
Vegas since the 1970s. From 1974 to 2010, the Clark County population grew by
over 580 percent, and major developments arose in the Las Vegas suburbs of
Summerlin, Green Valley, and Henderson. 22 Figure 11shows the extraordinary
growth of the Las Vegas area in satellite images from 1973 to 2006. 23 Moreover,
as Hughes et al (2008) note, factors such as Corporate Average Fuel Economy
(CAFE) standards, the growth of multiple income households and per capita
disposable income, and the evolution in public transit have changed the
responsiveness of U.S. consumers to changes in gasoline prices. Households may
also have shifted their beliefs about the extent to which changes in gasoline
prices are likely to persist.
22
23
http://www.lasvegasnevada.gov/files/PopulationEstimate.pdf
http://www.nasa.gov/images/content/202842main_lasvegas_lg.jpg
26
FIGURE 11. SATELLITE MAPS OF LAS VEGAS IN 1973, 1991, 2000, AND 2006
27
To see whether the pattern of estimated elasticities has changed substantially
over time, we conducted six hedonic regressions of overlapping 10-year subperiods of the data, inclusive of the years: 1976-85, 1981-90, 1986-95, 1991-2000,
1996-2005, and 2001-2010. The first six panels of Figure 12 show histograms of
the estimated price elasticities of home values with respect to gasoline prices in
those sub-periods of the data, following the estimation model reported in Table
3.
FIGURE 12. HISTOGRAMS OF PROPERTY PRICE-GAS PRICE ELASTICITIES, SUBPERIODS (VALUES FROM -0.5 TO 0.5 SHOWN)
PANEL 1: (1976-85); 2: (1981-90); 3: (1986-95); 4: (1991-2000); 5: (1996-2005); 6: (20012010)
PANEL 7: HOUSING BOOM, FROM JANUARY 2002 TO APRIL 2006 (INCLUSIVE)
PANEL 8: HOUSING BUST, FROM APRIL 2006 THROUGH DECEMBER 2010 (INCLUSIVE)
PANEL 9: FULL SAMPLE, FROM JANUARY 1976 THROUGH DECEMBER 2010
28
In the early periods, panels one through three, the elasticities are widely
dispersed, just shy of a uniform distribution. By the 1990s (panel four), the
elasticities begin to show a more strongly normal distribution centered between
0 and 0.1. By the last decade (panel six) of 2001 to 2010, the elasticities fall in a
much narrower distribution, with a density strongly concentrated in the 0 to 0.1
range. This suggests that the broad sensitivity (positive and negative) of home
prices to contemporaneous gasoline prices has fallen significantly over the period
of our data, and the distribution of elasticities has narrowed considerably.
The intensity of the relationship between housing values and gasoline prices may
have fallen in part as the overall fuel economy of the vehicle fleet has risen. For
example, the average fuel economy of new cars sold in 2010 was over 70
percent higher than new cars sold in 1975. 24 All else equal, we would expect this
to dampen the sensitivity of home prices to gasoline prices.
We also ran the regressions with data confined to the months of the Las Vegas
housing boom, as defined by the major run-up in the Case-Shiller housing price
index from January 2002 to its high in April 2006 (inclusive), and the housing
bust from April 2006 to the end of our data in December 2010 (inclusive). Panel
seven in Figure 12 shows the elasticities of sales in the months of the housing
boom. Panel eight includes sales in the months of the housing bust that followed.
The distribution of elasticities in the boom is similar to the almost uniform
distribution of earlier periods while elasticities in the bust are more narrowly
distributed with a mode near zero. This suggests that the distribution of
relationships (not just the average relationship) of gasoline prices to home values
could depend on the direction of the economy.
For comparison, panel nine in Figure 12 shows the elasticities of sales for the
entire period, the same data that appear in Figure 5. The spatial characteristics
(not shown) of the elasticities in later years (panels six through eight) indicate a
pattern of declining elasticities from the urban core slightly more pronounced
than elasticities in the full period shown in Figure 7. So although the distribution
of elasticities in later years is tighter, the overall spatial pattern of elasticities we
24
EPA (2013). Hughes et al (2008) find that the short-run price elasticity of U.S. gasoline demand
is significantly more inelastic in recent years than in the late 1970s.
29
see for the full sample holds throughout the period and may even intensify in
more recent years.
C. Comparative Statics
In this section, to assess the potential spread of home values produced by a
change in gasoline prices, we estimate the dollar value effect of an illustrative
increase of ten percent in gasoline prices on the relative value of homes in
different neighborhoods. 25
For comparison, according to the Energy Information Administration (EIA), H.R.
2454 (the Waxman-Markey climate bill) would have raised gasoline prices by 27
cents per gallon, or 7.6%, in 2020 relative the baseline value of $3.55 (in $2007).
Ten percent of the mean gasoline price in our data is about $0.22, a common
swing in gasoline prices, and it equates to the implications of a carbon tax of
about $22 per ton of CO2. 26 Such a tax is in the range of recent carbon tax
proposals and the social cost of carbon as estimated by the Obama
Administration. 27
To narrow the estimation results to plausible values, we windsorize the sample
of elasticities by dropping the highest and lowest two percent. To obtain the
change in home values relative to the mean, we subtract the mean elasticity
(about 0.06) from all of the estimated elasticities to obtain deviations in
elasticities from the mean. Then we multiply all of the deviations in elasticities by
ten to obtain the predicted percent change in home values for each tract given
an increase in gasoline prices of ten percent, relative to the mean. We evaluate
this change at the average property price in each tract to derive the expected
dollar level change in home prices, relative to the mean.
Figure 13 shows the graduated color map of the dollar value changes, relative to
the mean, for a ten percent ($0.22) increase in gasoline prices in central Clark
County. Figure 14 shows the results for the full county. The estimated home
price effects range from about -$7,850 to $5,590.
25
See EIA (2009), page 36, Figure 26.
Ramseur et al (2012), Table 3.
27
CBO (2013); U.S. Government Interagency Working Group on Social Cost of Carbon (2013)
26
30
FIGURE 13. ESTIMATED CHANGE IN PROPERTY PRICES WITH A 10% INCREASE IN
GASOLINE PRICES, RELATIVE TO MEAN, OUTLIERS DELETED
CENTRAL CLARK COUNTY (LAS VEGAS, NORTH LAS VEGAS, AND HENDERSON)
Central
Las Vegas
31
FIGURE 14. ESTIMATED CHANGE IN PROPERTY PRICES WITH A 10% INCREASE IN
GASOLINE PRICES, RELATIVE TO MEAN, CLARK COUNTY
The largest negative values are in the northern and southwest suburbs of the city,
and the largest positive values are in the urban core and the southern periphery.
These results reflect a combination of the magnitude of the elasticities and the
mean home values in these areas. Areas near major roads and highways, possibly
because they offer easy access to efficient travel paths, tend towards more
positive (blue) effects. That said, Figure 13 shows some adjacent neighborhoods
that have estimated price effects that are statistically significantly different from
zero and of opposite sign. This suggests there could be important unobservables,
perhaps such as demographics or access to public transit, that bear on the
results.
32
5.
CONCLUSION AND NEXT STEPS
We find that gasoline prices can affect property values in different
neighborhoods differently. Using data from 930,702 home sales in Clark County,
Nevada, from 1976 through 2010, we estimate the elasticities of property prices
with respect to gasoline prices, allowing the effect to vary by census tract. The
estimated elasticities range from about -.07 to about 0.18 across the census
tracts in the data, meaning that a ten percent increase in gasoline prices can shift
relative home values over a range about 2.5 percentage points.
These results suggest gasoline prices may be affecting credit risks, property
markets, and household wealth in ways the economic literature has so far not
fully recognized. While some studies have found that in general gasoline prices
are not significant determinants of home values and foreclosure, our results
suggest that there may be significant location-specific risks within metropolitan
areas. 28 We see some evidence of this in outlying suburbs but the effect is also
important in other areas.
Our investigation of the evolution of the relationship of gasoline prices to home
values suggests that the broad sensitivity (positive and negative) of home prices
to gasoline prices has fallen significantly since 1976. This could derive in part
from the 70 percent increase from 1975 to 2010 in the average fuel economy of
new cars sold in the United States.
Our results also bear on the potential distributional effects of a carbon tax or
other policy to price greenhouse gas emissions. A ten percent increase in the
gasoline price, a change on the order of the short run effects of recent proposed
climate legislation, is associated with changes in location-specific average home
values that span a range of over $13,000; households with homes near the
center of the city and near major corridors would be better off by up to about
$5,600, and some households living in the city outskirts would lower mean home
values by $7,800, relative to the mean. The net effect on much of the mid-city is
28
Molloy and Shan (2010), Duling (2008)
33
small, within $500 plus or minus. However, despite these broad spatial patterns
we find instances of significant differences in the effect of gasoline prices even
across adjacent neighborhoods. This suggests that location alone is not a perfect
predictor of how gasoline prices can affect home values and that there is more
to this story left for future investigation.
Our results could be pertinent to other important property markets. Cities with
housing supply constraints at least as binding as Las Vegas include, according to
Saiz (2010), New York, Los Angeles, Boston, Miami, and San Francisco.
Several extensions to our work are clear. First, given this evidence that gasoline
prices do matter, it would be useful to explore potential explanations of the
estimated elasticities other than location. For example, tracts with higher mean
household income may have property values that are less sensitive to gasoline
prices, positive or negative. This could have implications for the distribution by
income of the housing wealth effects of a carbon tax. Certainly commuting costs
as a share of home value are likely to vary by the overall price point of a
neighborhood. Other demographics, such as the share of non-working age adults,
might also matter. And tracts closer to employment centers, highways, and
public transport could systematically have relatively more positive relationship to
gasoline prices. Areas with older homes or greater population density may have
less elastic housing supply, and thus be more prone to gasoline price
capitalization (positive or negative). Relating the gasoline price effects to
neighborhood income levels would also estimate the potential impacts of carbon
pricing on households’ housing wealth by socioeconomic status.
Another question left open by this study is whether asymmetries in the
constraints on housing supply produce asymmetries in the effect of gasoline
prices. For instance, housing supply may be more elastic upward than downward,
particularly in city outskirts. Thus, outlying areas may experience larger negative
downward capitalization when gasoline prices rise than they do positive upward
capitalization when gasoline prices fall. Another finding worth investigating
further is the different patterns of elasticities in the months of a housing boom
and the months of the housing bust depicted in Figure 12.
34
Finally, a natural extension of our results would be to explore the spatial
patterns of the housing bubble collapse in Las Vegas starting in 2006, the worst
such implosion in the United States. 29 Cortright (2008) and Kaufman et al (2010)
suggest a role of energy prices in declining property market conditions in the
post-2005 era. Estimations could show whether Clark County homes in areas
with significant negative relationships between home values and gasoline prices
were especially prone to foreclosure.
REFERENCES
Alquist, R., and L. Kilian, "What Do We Learn from the Price of Crude Oil
Futures?" Journal of Applied Econometrics. 25. no. 4 (2010): 539–573.
Anderson, Soren T., Ryan Kellogg, and James M. Sallee, “What Do Consumers
Believe About Future Gasoline Prices?” NBER Working Paper No. 16974,
National Bureau of Economic Research, April 2011.
Carlos L. Araya, Trupti Kawli, Anshul Kundaje, Lixia Jiang, Beijing Wu, Dionne
Vafeados, Robert Terrell, Peter Weissdepp, Louis Gevirtzman, Daniel Mace,
Wei Niu, Alan P. Boyle, Dan Xie, Lijia Ma, John I. Murray, Valerie Reinke,
Robert H. Waterston, and Michael Snyder, “Regulatory analysis of the C.
elegans genome with spatiotemporal resolution ,” Nature 512,400–405(28
August 2014).
Bajari, P., and J. Cooley Fruehwirth, J, American Economic Review 2012, 102(5):
1898–1926
Bopp, A., and G. Lady. “A Comparison of Petroleum Futures Versus Spot Prices
as Predictors of Prices in the Future.” Energy Economics. 13. no. 4 (1991):
274-282.
Bureau of Land Management (BLM) Southern Nevada Public Land Management
Act of 1998, http://www.blm.gov/pgdata/content/nv/en/snplma.html
Busse, Meghan R., Christopher R. Knittel, and Florian Zettelmeyer. 2013. "Are
Consumers Myopic? Evidence from New and Used Car Purchases." American
Economic Review 103, no. 1: 220-256.
Campbell, J, S. Giglio, and P. Pathak. "Forced Sales and House Prices." American
Economic Review. 101. no. 5 (2011): 2108-2131.
Las Vegas experienced the worst year-to-year rate of home price decrease to December
2008 in the United States, according to the S&P/Case-Shiller home price index released February
24, 2009, as reported in Forbes: http://www.forbes.com/2009/02/24/housing-cities-ten-lifestylereal-estate_home_prices_slide_21.html?partner=msnre.
29
35
Chay, Ken and Michael Greenstone, “Does air quality matter? Evidence from the
housing market,” Journal of Political Economy, 113: 376-424.
Chinn, M, and O. Coibion. “The Predictive Content of Commodity Futures.”
Discussion paper. (2010). Williamsburg: College of William and Mary.
Clark County Department of Comprehensive Planning, Clark County Federal Lands
Report, May 22, 2013.
http://www.clarkcountynv.gov/Depts/comprehensive_planning/advanced_plan
ning/Documents/FederalLandsReport.pdf
Cortright, J. “Driven to the Brink: How the Gas Price Spike Popped the Housing
Bubble and Devalued the Suburbs.” Discussion paper. (2008). Chicago: CEOs
for Cities.
Cropper, M., L. Deck, and K. McConnell. “On the Choice of Functional Form for
Hedonic Price Functions.” Review of Economics and Statistics. 70. no. 4 (1988):
668-675.
Congressional Budget Office, Effects of a Carbon Tax on the Economy and the
Environment, Publication No. 4532, May 2013.
Coulson, N., and R Engle. “Transportation Costs and the Rent Gradient.” Journal
of Urban Economics. 21. no. 3 (1987): 287-297.
Dorsey, R., H. Hu, W. Mayer, and H. Wang
. “Hedonic Versus Repeat-Sales Housing Price Indexes for Measuring the Recent
Boom-Bust Cycle.” Journal of Housing Economics. 19. no. 2 (2010): 75-93 or
87-105.
Duling, Bryan. “The determinants of foreclosures for single-family homes in the
United States.” 2008. The Park Place Economist. 16 45-56
Eichholtz, Piet, Nils Kok, and John M. Quigley. 2013. "The Economics of Green
Building." Review Of Economics And Statistics 95, no. 1: 50-63.
Environmental Protection Agency (EPA). “Supplemental EPA Analysis of the
American Clean Energy and Security Act of 2009 H.R. 2454 in the 111th
Congress.” (2010).
Environmental Protection Agency (EPA). “Trends Report: Light-Duty
Automotive Technology, Carbon Dioxide Emissions, and Fuel Economy
Trends: 1975 Through 2013,” EPA-420-R-13-011, (2013).
Gayer, Ted, James T. Hamilton, W. Kip Viscusi, “The Market Value of Reducing
Cancer Risk: Hedonic Housing Prices with Changing Information,” Southern
Economic Journal, Vol 69, No. 2 (October 2002), pp. 266-289.
Glaeser, Edward L., Joshua D. Gottlieb, and Kristina Tobio. "Housing Booms and
City Centers." American Economic Review 102, no. 3 (May 2012): 127-133.
Goodman, A, and T Thibodeau. “Dwelling-Age-Related Heteroskedasticity in
Hedonic House Price Equations: An Extension.” Journal of Housing Research. 8.
no. 2 (1997): 299-317.
36
Halvorsen, R., and H. Pollakowski. “The Effects of Fuel Prices on House Prices.”
Urban Studies. 18. no. 2 (1981): 205-211.
Hassett, Kevin, Aparna Mathur, and Gilbert Metcalf, “The Incidence of a U.S.
Carbon tax: a Lifetime and Regional Analysis,” Vol. 30, No. 2 of The Energy
Journal (April 2009), pp. 155-178.
Hughes, J. E., Knittel, C. R., & Sperling, D. (2008). Evidence of a Shift in the
Short-Run Price Elasticity of Gasoline Demand. Energy Journal, 29(1), 113134.
Kaufmann, Robert K., Nancy Gonzalez, Thomas A. Nickerson, Tyler S. Nesbit,
“Do household energy expenditures affect mortgage delinquency rates?”
Energy Economics 33 (2011) 188–194.
Langer, A., and N. H. Miller (2013). Automakers' Short-Run Responses to
Changing Gasoline Prices. Review of Economics and Statistics, 95(4), 1198-1211.
Molloy, Raven, and Hui Shan. "The Effect of Gasoline Prices on Household
Location." Review of Economics and Statistics, October 10, 2013, 95(4): 12121221.
Mathur, Aparna, and Adele Morris, “Distributional Effects of a Carbon Tax in
Broader U.S. Fiscal Reform,” Energy Policy 66 (2014), pp. 326-334.
Nichols and Schaffer (2007), “Clustered Errors in Stata,” presentation,
September 10, 2007. http://www.stata.com/meeting/13uk/nichols_crse.pdf
Padmos RC, Schloot NC, Beyan H, Ruwhof C, Staal FJ, de Ridder D, Aanstoot HJ,
Lam-Tse WK, de Wit H, de Herder C, Drexhage RC, Menart B, Leslie RD,
Drexhage HA, “Distinct Monocyte Gene-Expression Profiles in Autoimmune
Diabetes,” Diabetes. 2008 Oct; 57(10):2768-73
Ramseur, Jonathan, Jane Leggett, and Molly Sherlock. Carbon Tax: Deficit Reduction
and Other Considerations. CRS Report for Congress R42731. Washington, DC:
Congressional Research Service, 2012.
Rausch, S., G.E. Metcalf, and J. M. Reilly. 2011. Distributional Impacts of Carbon
Pricing: A General Equilibrium Approach with Micro-Data for Households.
Energy Economics 33: S20–S33.
Rosen, Sherwin, Hedonic prices and implicit markets: Product differentiation in
pure competition, Journal of Political Economy, 82:34-56,1974.
Saiz, Albert, “The Geographic Determinants of Housing Supply, The Quarterly
Journal of Economics (2010) 125 (3): 1253-1296
Sexton, S., J. Wu, and D. Zilberman. “ How High Gas Prices Triggered the
Housing Crisis: Theory and Empirical Evidence.” Discussion paper. (2012).
Berkeley: UC Center for Energy and Environmental Economics.
Bureau of Land Management (BLM) Southern Nevada Public Land Management
Act of 1998, http://www.blm.gov/pgdata/content/nv/en/snplma.html
U.S. Energy Information Administration, Energy Market and Economic Impacts of
H.R. 2454, the American Clean Energy and Security Act of 2009, August 2009.
37
U.S. Government Interagency Working Group on Social Cost of Carbon,
Technical Support Document: Technical Update of the Social Cost of Carbon for
Regulatory Impact Analysis Under Executive Order 12866, May 2013.
Walls, M., K. Palmer, and T. Gerarden. Is Energy Efficiency Capitalized into
Home Prices? Evidence from Three U.S. Cities.” Discussion paper (2013)
Washington, D. C., Resources for the Future.
38
APPENDIX
A. Data discussion
This section discusses the construction of the dataset used in the regressions.
This study combines data from three separate data sources. For transaction
information, the study uses data from home sales in Clark County, Nevada, from
January 1976 to December 2010, obtained from the Clark County Assessor’s
Office (CCAO). 30 We use the x and y coordinates to map every property to the
most recent available census tract boundaries. Thus we ensure that the tract
identifier for each property is constant for all the sales of the property in our
data, even if the official census tract of the property has changed over time.
The geographic distribution by census tract of the observations in the dataset
appears in Figure A. Panel 1 shows that the census tracts with the greatest
numbers of property sales in our data are concentrated in central Clark County
near the cities of Las Vegas, Henderson, and North Las Vegas. Each dot in the
inset of Panel 1 represents the location of a transaction, indicating that
transactions within large tracts are generally spatially concentrated in a small
area of the tract. Panel 2 shows that within central Clark County, the census
tracts with the largest numbers of transactions in our data fall outside the city
center of Las Vegas, indicated by the diagonal white and light pink areas in the
center left of Panel 2.
30
Further documentation on the data set is available at the Clark County Assessor’s website.
Information about the residential records appears at:
http://www.clarkcountynv.gov/Depts/assessor/Pages/ResidentialRecordLayoutInformation.aspx.
Information about the specific sales codes appears at:
http://www.clarkcountynv.gov/Depts/assessor/Documents/Sales%20Codes.pdf.
39
FIGURE 15. GEOGRAPHIC DISTRIBUTION OF PROPERTY TRANSACTIONS:
PANEL 1: ALL OF CLARK COUNTY, NUMBER AND LOCATION OF TRANSACTIONS
40
PANEL 2. -- CENTRAL CLARK COUNTY
We exclude observations that are unlikely to represent ordinary arms’ length
transactions or vacant land. To do this, we use the CCAO codes that indicate
the type of sale for improved properties. Sale types included in the dataset are
listed in Table A.1.
41
TABLE A.1. CCAO SALES CODES INCLUDED IN RAW DATA
(975,584 OBSERVATIONS)
Sale Code
R
Y
Z
T
N
X
W
M
Description
Regular sale
Resale in the market range
First time sale in the market range
Contract sale, includes land and new construction
First time sale or resale above or below market, includes
sacrifice, second mortgage, introduction sales
Foreclosure resale, tax sale, estate sale in the market range
Foreclosure resale, tax sale, estate sale above or below
market
Sales price is the total purchase price for many or numerous
parcels being transferred on the same deed or one purchase
price was paid for several parcels conveyed on several deeds.
We constructed a separate foreclosure indicator using the CCAO Foreclosure
code. We also create indicator variables for the years in which the transaction
occurs and convert all price variables to real 2010 dollars using the Consumer
Price Index (All Urban Consumers: U.S. city average deflator) from the U.S.
Bureau of Labor Statistics. 31
We dropped observations with values that suggest the property is likely to be
extremely unusual, a non-arms’-length transaction, or coded in error. Thus we
drop properties with a real price of less than $40,000 or less than one full bath.
Given the specialized market for such properties, we exclude homes on lots
larger than five acres and greater than 8,000 square feet of living space. We also
drop sales of homes with prices over $5 million ($2010) or with more than six
full baths. We also drop properties with an unusually high price per square foot
of over $300 per square foot. The deleted observations collectively represent
4.6 percent of the raw data set.
31
http://www.bls.gov/cpi/cpid10av.pdf. BLS Table 1A.
42
Using the year and month of the transaction, we merge the property data with
data for the national average gasoline price and the Nevada state unemployment
rate. We created logged values for continuous variables, including the sales price
of the property, gasoline prices, living area, and lot size. To compute the age of
the home in years, we subtract the year of the transaction from the variable in
the data that indicates the year the home was built. In some cases the property
is sold before the home is built, for example through an advance property sale.
In that case we set the age to zero. Home sales with an age of zero appear as
“New” in the blue bars of FIGURE 2.
B. Repeat sales alternative model
A repeat sales approach eliminates the problem of omitted variables with
respect to time-invariant characteristics of the property. Assuming the changes
in housing prices are in percentage terms, we write this log-log model:
1)
Δyist
= αs + γt + λsΔgmαs + μsΔum αs + ϵist.
The dependent variable is the difference in the log price of a home from one sale
to the next, Δyist. This is the appreciation of the price of property i in census tract
s over the period between sales. The variable ϵist is an error term that reflects
random variation in house prices, and the terms αs are indicators for the census
tracts in which the properties lie. The year indicators γt in the repeat sales
model are a little different than in the hedonic model because we want to
control for year effects both in the year of purchase and sale. Thus γt is -1 if the
home was purchased in year t and +1 if it was sold in year t. This approach
ensures that the year-specific effects on home price appreciation are symmetric
but of opposite sign, depending on whether the transaction was a purchase or a
sale.
The terms Δgmαs are the change in real log gasoline prices from purchase to sale,
interacted with census tract indicators. Likewise, the terms μsΔum αs interact the
change in state unemployment levels and census tract indicators. Our data
report almost no appreciable changes in property characteristics, so Equation 2
does not include a vector of changes in characteristics corresponding to the β′ Xi
in Equation 1. The estimated elasticities of property prices with respect to
43
gasoline prices in the repeat sales model (Equation 2) will not match perfectly
the estimated elasticities in the hedonic model (Equation 1). Although
arithmetically the two coefficients are the same, only about 442,900 of the
approximately 931,000 homes sales in the full dataset are repeat sales, i.e. sales
of a property for which we also have data on a prior sale. For example, the
repeat sales data exclude all sales of new homes (except as a cost basis for
subsequent sales), represented by the blue bars of FIGURE 2, along with the first
sale of each non-new property after December 1975. Table A.2 reports the
repeat sales data summary statistics and estimation model.
TABLE A.2.− REPEAT SALES MODEL
SUMMARY STATISTICS AND ESTIMATION MODEL
Number of Observations = 442,900
Mean
∆ln(Property price)
-.047
Standard
Deviation
.484
Min
Max
-2.90
3.46
∆ln(Price per gallon of
gasoline)
.098
.288
-1.04
1.21
∆Nevada state-level
unemployment rate
(percentage points)
1.49
3.70
-8.4
11.1
DEPENDENT VARIABLE: CHANGE IN LN(PROPERTY SALE PRICE) SINCE LAST SALE
Year indicators for 1977 through 2010 (= -1 if the home was purchased
in year t and +1 if it was sold in year t)
Census tract indicators for 335 tracts
Census tract indicators interacted with change ln(gasoline price) from
purchase to sale
Census tract indicators interacted with change in Nevada
unemployment rate from purchase month to sale month
Number of observations =
R-squared =
Root MSE =
44
Yes
Yes
Yes
Yes
442,900
0. 644
0.290
FIGURE 16 below shows histogram of the estimated elasticities for the repeat sales
model. The repeat sales results show a pattern broadly similar to that in Figure 5
for the hedonic model, but with a somewhat greater density in the positive range
between 0 and 0.25. This is consistent with the results in Figure 12, panels 5 and
6, which show a generally more positive set of elasticities in the later decades of
the data.
FIGURE 16. ESTIMATED ELASTICITIES OF PROPERTY PRICES WITH RESPECT TO GASOLINE
PRICES: REPEAT SALES AND HEDONIC MODELS
Figure 17 below is a graduated color map of the results from the repeat sales
model. The map shows an overall pattern of larger positive elasticities in the
urban core and elasticities that are closer to zero or slightly negative in the
outskirts. The pattern of declining elasticity from the urban core is more
pronounced in the repeat sales map than it is in Figure 7, which maps the results
from the hedonic model estimated with the full dataset. This could be due to
either the improved control for location inherent in a repeat sales approach, but
it is also consistent with the pattern in more recent property sales as discussed
per Figure 12.
45
FIGURE 17. REPEAT SALE MODEL
ESTIMATED ELASTICITIES OF PROPERTY PRICES
WITH RESPECT TO GASOLINE PRICES:
CENTRAL CLARK COUNTY (LAS VEGAS, NORTH LAS VEGAS, AND HENDERSON)
Central
Las Vegas
46